AI Scam Uses Fake NZ Nana Raewyn to Sell Temu Baskets

An ad featuring 68-year-old Raewyn Thorburn, an Otago nana selling her final handmade baskets before retirement, turned out to be more than meets the eye.

The heart-tugging backstory vanished under scrutiny of her American lilt and unusual pronunciation of the region’s name. Raewyn never existed, and neither did her workshop.

Social media users have shared warnings after discovering Raewyn also hawked musical instruments under the name Maree. That identical stock appears on Temu for a fifth of the price.

Detecting Fake Advertisements

Dr Andrew Lensen, a senior lecturer in artificial intelligence at Victoria University, watched the ad several times. He spotted new tell-tale signs with each rewatch.

“We see that quite often in these sorts of formats because when you create a video, you can only make a short one – a few seconds,” Lensen said.

“You can see the background changes in different scenes; the captions don’t quite line up – those sorts of things. And it’s sort of oversaturated as well, so there’s lots of small signs but the voice, obviously, was the main one for sure.”

Background shots shift unpredictably across scenes. Captions miss exact audio synchronization, while colors carry an oversaturated finish.

Lower Barriers to Entry

Fictional salespeople complete with custom voices and fabricated workshops now require minimal technical skill. Online services produce these synthetic presenters quickly and cheaply.

“A few years ago, you could do it, but you had to have a lot of skill, a lot of expertise, a lot of money or resources,” Lensen said.

“Whereas now, there’s services online that we can almost sell this to you at quite a cheap price and so the barrier to entry’s a lot lower.”

Operators can generate a new persona after a few days of training. They then establish automated workflows to repeat the process indefinitely.

Independent Verification Needed

Visual clues like deformed hands remain unreliable indicators as generative models improve. Accents require more training data due to a shortage of local voices.

“So, when you go and look at this product they’re trying to sell, can you find reviews of it? Can you do an image search and see where else it’s listed?” Lensen said.

“In this case, it was also on Temu, I think, so those sort of critical thinking, independent checks are probably the way forward.”

Major platforms maintain detection tools yet leave unverified content active. Lensen noted the Raewyn ad remained visible despite numerous user reports.

“I can still see it on there even know I know many, many people have reported it, so there’s a big question there as to why they’re not taking it down, even when they’re being told about it,” Lensen said.

Frequently Asked Questions

How can shoppers identify synthetic product videos?

Look for inconsistent background details, oversaturated visuals, captions that do not align with the audio, and cloned voices with mismatched regional accents. Independent image searches reveal if identical items sell elsewhere for lower prices.

Why do major platforms keep flagged ads online?

Dr Andrew Lensen said that platforms like YouTube have detection forms in place but often fail to remove reported synthetic ads promptly, leaving users unprotected and platforms unaccountable.

Andrew Lensen, a senior lecturer in artificial intelligence at Victoria University of Wellington, received the 2026 Cooper Award from Royal Society Te Apārangi for his work on making AI more transparent and trustworthy, and a 2026 Critic and Conscience of Society award. He serves as the Programme Director of Artificial Intelligence at the university, which offers the first AI major in New Zealand.

Are local accents difficult for artificial intelligence to replicate?

Accents take longer for models to perfect because training datasets contain limited local voice samples, though the technology improves rapidly.